Biomarkers of food intake
The limitations associated with traditional methods for assessment of food intake are well established(Reference Park, Dodd and Kipnis1–Reference Prentice, Tinker and Huang4). Examples of such limitations include under-reporting, poor estimation of portion size and recall errors. Biomarkers of food intake have the potential to be objective measures of intake thus addressing some of the limitations of self-reported traditional methods. The current review will examine the field of food intake biomarkers and the potential use of such biomarkers in nutrition research. Food intake biomarkers can reflect food intake and are exogenous metabolites originating in the food(Reference Gao, Pratico and Scalbert5). While food intake will impact endogenous metabolites (metabolites synthesised by human metabolic pathways), these metabolites are not robust food intake biomarkers(Reference Cuparencu, Bulmus-Tuccar and Stanstrup6). Instead, these endogenous metabolites can support the evidence base for the impact of certain foods and dietary patterns on metabolic health(Reference Brennan and Hu7). To advance the nutrition assessment field food intake biomarkers should also reveal quantitative information about food intake. While many studies have emerged in the literature with lists of biomarkers of intake for certain foods caution needs to be exerted when interpreting these. Many still contain endogenous metabolites and very few have been assessed against validation criteria.
As part of a European Joint Programming Initiative, the FoodBall consortium a series of systematic literature reviews were conducted to examine biomarkers of intake for a range of foods including for example citrus, red meat, coffee, green leafy vegetables, cereal foods, apple, pear and stone fruit(Reference Ulaszewska, Garcia-Aloy and Vazquez-Manjarrez8–Reference Munger, Garcia-Aloy and Vazquez-Fresno13). A common trend that emerged from these is that many putative biomarkers of intake are not well validated in terms of the validation criteria for biomarkers of food intake(Reference Dragsted, Gao and Scalbert14). A more recent assessment of these in terms of some of the validation criteria also revealed that many of the foods are without well-validated biomarkers(Reference Cuparencu, Bulmus-Tuccar and Stanstrup6). Independent of this another review reported 347 potential biomarkers for 67 foods with biomarkers for wholegrains, soy and sugar being the most reliable(Reference Clarke, Rollo and Pezdirc15). The present review will focus on the use of such biomarkers in Nutrition Research and the reader is referred to the above systematic reviews for details on biomarkers for specific foods.
Discovery of food intake biomarkers
The preferred approach for identification of biomarkers of food intake is to employ a human intervention study design. Typically, this involves participants consuming specific foods with collection of biological samples in the post prandial state(Reference Brouwer-Brolsma, Brennan and Drevon16–Reference Vazquez-Manjarrez, Weinert and Ulaszewska19). Biological samples can include blood and urine and the exact time points collected will vary with the food in question. Collection of samples up to 24 h and 48 h post consumption of the food can be useful to assess the half-life of the biomarkers. While acute post prandial studies are common approaches for the discovery of food intake biomarkers it is also possible to use other timeframes for the interventions. For example, short-term studies including consumption of foods for days or weeks are also viable options(Reference Saenger, Hubner and Lindemann20). A key consideration in all these interventions is the inclusion of a control arm with consumption of the control food: this helps to ensure that the biomarkers emerging are specific to the food of interest and should not be changing in the control food. A recent study employed an approach of supplying the habitual diet to 153 individuals over a 2-week period(Reference Playdon, Tinker and Prentice21). Consequently diet-metabolites associations were identified for 23 foods, beverages and supplements.
Many studies have also employed observational data to identify biomarkers that are associated with dietary factors estimated from self-reported tools such as Food Frequency Questionnaires (FFQ) or 24-hour recalls(Reference Noerman, Johansson and Shi22–Reference Garcia-Gavilan, Babio and Toledo24). When employing this approach caution is needed as there is a high risk of confounding as foods are consumed in patterns and dissociating an association of a specific food with a biomarker from associations with other foods can be difficult. Furthermore, many of the metabolomic techniques used in such studies cover predominantly endogenous metabolites thus limiting the ability to identify biomarkers that will satisfy validation criteria. Notwithstanding these issues, there are valid uses for observational data in the field: such data can provide important confirmation of relationships observed in smaller intervention studies and allow the possibility of examination of confounding factors such as age, sex and BMI.
Validation of food intake biomarkers
While there is still a focus on identifying food intake biomarkers the validation of such biomarkers has lagged behind. In an effort to progress the field, the European FoodBall project proposed a serious of validation criteria including plausibility, dose-response, time-response, robustness, reliability, stability, analytical performance and reproducibility(Reference Dragsted, Gao and Scalbert14). More recently an extra condition was added: assessment of the intra and inter-individual variability in the biomarker levels(Reference Cuparencu, Bulmus-Tuccar and Stanstrup6). Assessment of biomarkers against these criteria should help biomarkers that will have high utility in the field. Assessing the plausibility of a food intake biomarker involves verifying its specificity to the food and identifying any food chemistry, processing, or experimental factors that could explain the increased concentration after consumption. The response of the food intake biomarker to varying portions of a specific food should be evaluated, considering factors such as intake range, habitual baseline levels, bioavailability, excretion kinetics and saturation thresholds. The kinetics of excretion of the biomarker following consumption of the food is important to understand and ideally, the half-life of the biomarkers should be established. To enable use in nutrition it is important that any biomarkers are robust across a number of population groups and have limited interactions with other foods. In the context of food intake biomarkers, reliability refers to the agreement with other biomarkers or self-reported assessment methods. If no other biomarkers exist, the comparison with self-reported data can be challenging in the sense that the biomarkers are being developed due to limitations associated with the self-reported data. Expecting good agreement for all foods is not realistic and some flexibility is needed when assessing the reliability of biomarkers against self-reported data. The validation criteria stability, analytical performance and reproducibility relate to the chemical stability of the biomarker and the analytical procedures for measurement. For effective use in nutrition research, food intake biomarkers must remain stable in the biofluid to be used for the analysis. The biomarker’s analytical performance should be thoroughly documented, including its precision, accuracy, detection limits, and assessment of inter- and intra-batch variation. The biomarker’s performance and efficacy should be reproducible, with, if possible, data to demonstrate the consistency of results across different laboratories. An example of a biomarker that has been extensively validated is proline betaine: using different analytical techniques in various labs it was demonstrated to distinguish between low, medium and high consumers(Reference Saenger, Hubner and Lindemann20,Reference Gibbons, Michielsen and Rundle25) . Furthermore, in an observational study, good agreement with a 7-day food record was reported(Reference Hu, Wang and Willett26).
As mentioned above an additional criterion of variability has been proposed recently(Reference Cuparencu, Bulmus-Tuccar and Stanstrup6). It is of course desirable that biomarkers have low variability within and between individuals. Assessment of the variability of the biomarkers can be performed with repeated measurements. Furthermore, rigorous statistical analysis in the initial stages of biomarker discovery is likely to remove biomarkers with large inter-individual variation. Overall, validation is a key part of ensuring that biomarkers of food intake are suitable for the desired purpose and have the potential to make an impact in the field of nutrition.
Important considerations for food intake biomarkers
Many food intake biomarkers reflect short-term intake and to achieve an assessment of habitual dietary intake it is important to use repeated measures of biomarker levels from multiple biological samples taken over a timeframe(Reference Saenger, Hubner and Lindemann20,Reference Beckmann, Wilson and Lloyd27) . Considering the guidance for 24 h recalls recommends to capture data in at least 2 non-consecutive days within a maximum timeframe of between 2 and 6 weeks, a similar approach could be applied to biomarkers. Indeed three 24-hour urine samples collected over time reflected long-term intake of sweeteners and certain polyphenol-containing foods(Reference Sun, Bertrand and Franke28,Reference Buso, Boshuizen and Naomi29) . Analysis of polyphenol biomarkers (for example ferulic acid, kaempferol and hesperetin) indicated good reproducibility and that three 24-h urine samples would have been sufficient to achieve a Reliability Index of 0·8(Reference Sun, Bertrand and Franke28).
Another consideration is the type of biofluid to be used. With respect to urine samples, recent data has indicated that spot urine samples work well for measurements of biomarkers of food intake(Reference Beckmann, Wilson and Lloyd27) supporting the use of spot urine samples in lieu of the more burdensome 24-hour urine samples. Examples of biomarkers working well in spot urines include but are not limited to proline betaine for citrus intake, trigonelline for legume intake and tartaric acid for grape intake(Reference Beckmann, Wilson and Lloyd27). It is important to acknowledge that the majority of the work to date with respect to food intake biomarkers has focused on urine as the biofluid. However, it is likely that our current routine methods of measuring metabolomics in plasma/serum samples are likely not to be geared towards measurement of the food-derived compounds. Further sample treatment and removal of abundant endogenous molecules prior to metabolomics analysis should be explored for discovery of food intake biomarkers in plasma and serum. Other biofluids have also shown promise and further work is needed to examine their full potential. Included in this are saliva and hair(Reference Votruba, Shaw and Oh30). In particular, the naturally occurring carbon and nitrogen stable isotope ratios in hair have been proposed as a means of assessing added sugar intake and sugar-sweetened beverage intake(Reference Nash, Kristal and Hopkins31,Reference Ko, O’Brien and Rivera32) . A recent review highlighted the potential of these approaches but also identified key issues that need to be addressed for progress to be made including the need to demonstrate the ability of such approaches to be responsive to intervention studies(Reference Tripicchio, Smethers and Johnson33).
Implementing food intake biomarkers in nutrition research
Despite challenges with the discovery of biomarkers of food intake, there are some excellent examples of how to use them in nutrition research. While still in its infancy there are noteworthy examples of utility that illustrate the potential of such biomarkers (Figure 1). Biomarkers can be used as measures of adherence to dietary interventions: implementation of biomarkers to demonstrate adherence has the potential to enhance randomised controlled nutrition interventions. Compliance to diets is often called into question in nutrition studies. A recent study of the Anti-inflammatory Diet In Rheumatoid arthritis trial indicated that assessment of both biomarkers and self-reported data revealed good compliance to the study diets with respect to wholegrains cooking fat, seafood and red meat(Reference Wadell, Barebring and Hulander34). However, compliance to the study with respect to fruit and vegetables was uncertain. This study highlights the potential for assessment of compliance/adherence to diets and can aid in the interpretation of study outcomes.
Fig. 1. Overview of some applications of food intake biomarkers in nutrition research.
One of the main hopes for food intake biomarkers is that they will provide accurate information on the quantity of consumption of a specific food. However, despite the growing number of proposed biomarkers for food intake, there are still limited examples demonstrating their ability to quantify food intake accurately. In our previous work, we demonstrated that proline betaine as a biomarker of citrus intake could accurately predict intake both in an intervention setting and a free-living environment(Reference Gibbons, Michielsen and Rundle25). In a separate study, urinary tartaric acid demonstrated a good ability to estimate grape intake with strong agreement between estimated and actual intake, with a reported correlation coefficient of R2 = 0·9(Reference Garcia-Perez, Posma and Chambers35). Other examples have demonstrated the ability of biomarkers or panels of biomarkers to classify individuals into different categories of consumption(Reference Vazquez-Manjarrez, Weinert and Ulaszewska19,Reference Garcia-Aloy, Llorach and Urpi-Sarda36,Reference Xi, Berendsen and Ernst37) . A combination of 2 biomarkers (methoxyeugenol glucuronide and dopamine sulfate) could classify individuals into high and non-consumers of banana(Reference Vazquez-Manjarrez, Weinert and Ulaszewska19). With respect to walnut consumption, a combination of 18 metabolites could classify individuals into non-consumers of walnut or habitual walnut consumers(Reference Garcia-Aloy, Llorach and Urpi-Sarda36). For coffee biomarkers, a combination of 16 biomarkers could classify into low consumers or high consumers(Reference Xi, Berendsen and Ernst37).
Examples of using food intake biomarkers to calibrate self-reported dietary data have emerged in recent years. Using a combination of urinary and serum biomarkers for red meat intake calibration equations were developed with data from 450 participants(Reference Zheng, Pettinger and Gowda38). These calibration equations were then used to adjust the FFQ data in a larger study to examine associations between red meat intake and certain diseases. The employment of the biomarkers enabled the authors to conclude that while the higher meat intake was associated with increased chronic disease risks the associations were attributable to higher fat, energy and sodium intake as opposed to the meat intake(Reference Zheng, Pettinger and Gowda38). Other studies have also employed this approach and demonstrated stronger associations with disease risk and outcomes(Reference Prentice, Pettinger and Neuhouser39–Reference Prentice, Vasan and Tinker41). For example, biomarker calibration of Health Eating Index scores resulted in lower risks for cancer, T2DM and mortality(Reference Neuhouser, Pettinger and Tinker42). Our own work has also demonstrated the potential of correction of self-reported data for specific food intake using biomarkers. Using proline betaine as a biomarker of citrus intake we developed calibration equations that could be used to correct self-reported data when only self-reported data is available(Reference D’Angelo, Gormley and McNulty43). Overall, these studies demonstrate clearly the potential of food intake biomarkers in improving the overall accuracy of dietary assessment. Further work is needed to expand the range of biomarkers to cover more foods of relevance for health and disease.
Not all food intake biomarkers are capable of delivering quantitative information on food intake. However, such biomarkers may have potentially important applications in terms of studying mechanistic pathways between food intake and disease outcomes. For example, it is well established that diet impacts the formation of short-chain fatty acids by the gut microbiota. A recent study demonstrated that the short-chain fatty acids may be key mediators for the relationship between the Mediterranean diet and gut health(Reference Seethaler, Nguyen and Basrai44). In this study, the short-chain fatty acids mediated the effect of the Mediterranean diet on intestinal barrier function in the 3 months intervention. In a prospective examination of data from the UK biobank, a series of metabolites were identified as mediating the association between dietary pattern scores and overall cancer risk(Reference Fan, Hu and Xie45). These and other examples illustrate the potential of biomarkers influenced by food intake to enhance our understanding of the relationships between diet and disease development. Furthermore, these biomarkers may also play a role in the development of targeted interventions for disease prevention.
Concomitant with the increased interest in assessment of dietary patterns in nutrition literature a number of studies emerged where combinations of metabolites (metabolites scores) were developed for adherence to dietary patterns(Reference Li, Guasch-Ferre and Chung46–Reference Kim, Anderson and Hu48). It is worth noting the majority of these metabolites encompass endogenous metabolites and reflect impact of adherence on certain metabolic pathways. Nonetheless, these studies have revealed important insights. For example, a combination of 67 metabolites associated with adherence to the Mediterranean diet was predictive of Cardiovascular Disease (CVD) risk in a prospective analysis(Reference Li, Guasch-Ferre and Chung46). A smaller number of metabolites (8 in total) were inversely associated with 3 healthy dietary patterns and associated with poorer cardiometabolic traits and increased diabetes risk(Reference Chen, Chai and Xing49). While these studies are not employing solely biomarkers that are food intake biomarkers the concepts and results can be useful for the development of precision nutrition and the metabolites can be viewed as useful targets for the development of targeted interventions. For advancement of dietary assessment approaches it would be useful to assign individuals to dietary patterns using biomarker data only. While further work is needed in terms of discovery and validation of food intake biomarkers there is some existing evidence that this will be possible. Using data from a controlled intervention a model was built based on urinary metabolomics data that classified participants into dietary patterns. The model was validated in two separate population groups(Reference Garcia-Perez, Posma and Gibson50). Work emerging from our research group showed that it was possible to classify individuals into one of 4 dietary patterns with good reproducibility over 4 timepoints using only urinary biomarker profiles(Reference Prendiville, Walton and Flynn51). Overall, while significant effort is directed towards identifying biomarkers of food intake it is imperative that the field also demonstrates how we can use such biomarkers. Furthermore, demonstrating the added value and utility in nutrition research will be key for growth and development of the biomarker field.
Challenges and future directions
While the field of food intake biomarkers has gained considerable attention in recent years there are a number of challenges that remain. Firstly it is important to re-emphasise that many metabolomics platforms are established to measure endogenous metabolites and do not cover metabolites that would be classified as food intake biomarkers(Reference Cuparencu, Bulmus-Tuccar and Stanstrup6). Furthermore, the metabolome is highly dynamic with many factors influencing it. Acknowledging this and encompassing it in the interpretation of the data will be important to develop the field further.
One of the key challenges currently in the discovery and development of food intake biomarkers is the lack of databases with food-derived biomarkers. This makes metabolite assignment difficult and time-consuming. Further development of databases such as FoodDB would help accelerate this aspect of the work. For this international alignment and sharing of spectral data and standards would advance the field. foodMASST was recently released and can enable spectral matching of both known and unknown features in foods and drinks(Reference West, Schmid and Gauglitz52). Work from the Periodic Table of Food Initiative will deliver important information on the metabolite composition of food(Reference Ahmed, de la Parra and Elouafi53). For example, having available spectra from a range of foods will help in assessing the plausibility of biomarkers identified in human biological samples following consumption of the food. Despite these advancements, there is still the need for development of databases that enable sharing of spectral data from biological samples and standards of potential interest. International efforts that can address this will enable the field to advance.
While the discovery of food intake biomarkers has gained attention in recent years there is now a need to move forward and demonstrate the utility in terms of assessment of food intake. In this respect, there have been some encouraging examples as indicated above. For application to be successful validation of food intake biomarkers is key. Further work is also needed to ensure that the biomarkers are valid in different population groups. In large epidemiological studies combining biomarkers with self-reported data has great potential. However, further work is needed to expand the range of biomarkers to ensure that the key foods related to h`ealth are covered. The majority of current food intake biomarkers reflect short-term intake and new approaches for repeated collection and measurement of biological samples are needed. Conclusion
Significant progress has been made in recent years in the field of food intake biomarkers. The majority of work to date has focused on analytical and discovery aspects. There is now a need to move forward with the applications of such biomarkers in nutrition research. The present review illustrates some key ways we can use these biomarkers while also acknowledging the challenges in the field. An area of particular interest for the nutrition community is the work on combining dietary and biomarker data to improve assessment of dietary intake. Addressing the challenges in the field with International efforts should enable significant advancement of the field.
Financial support
The author acknowledges support from the Health Research Board Ireland for the following grants: US-Tri-partite grants USIRL-2019-1 and JPI-HDHL-2021-1.
Competing of interests
The author(s) declare none
Open access
